| Summary: | 碩士 === 國立中興大學 === 資訊科學與工程學系所 === 100 === The neural trees (NTs) and their variants recently have been proposed. In a NT, each internal node contains a neural network. That is, the input samples can be classified in a NT by these neural networks. In general, the performance of NTs outperforms the decision trees (DTs). The reason is that DTs are only suitable to classify the samples in which the distribution of samples is linear. However, NTs can classify the samples in which the distribution of samples is nonlinear. In NTs, the neural network is designed by the back-propagation method. The disadvantage of the back-propagation method is that it can easily search for the local minimal solution. In this thesis, the genetic algorithm (GA) is proposed to replace the back-propagation method to design the neural network. Therefore, the proposed NT is called the genetic neural tree (GNT) in this thesis. Also, the growing method is proposed to design the GNT according to the recognition rate and computing complexity of GNT. Thus, the GNT is near-optimal.
|
|---|
